1 /*
2 * Copyright 2009 VMware, Inc.
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 *
22 * Authors: Michel Dänzer
23 */
24 #include <drm/drmP.h>
25 #include <drm/radeon_drm.h>
26 #include "radeon_reg.h"
27 #include "radeon.h"
28
29 #define RADEON_TEST_COPY_BLIT 1
30 #define RADEON_TEST_COPY_DMA 0
31
32
33 /* Test BO GTT->VRAM and VRAM->GTT GPU copies across the whole GTT aperture */
radeon_do_test_moves(struct radeon_device * rdev,int flag)34 static void radeon_do_test_moves(struct radeon_device *rdev, int flag)
35 {
36 struct radeon_bo *vram_obj = NULL;
37 struct radeon_bo **gtt_obj = NULL;
38 struct radeon_fence *fence = NULL;
39 uint64_t gtt_addr, vram_addr;
40 unsigned i, n, size;
41 int r, ring;
42
43 switch (flag) {
44 case RADEON_TEST_COPY_DMA:
45 ring = radeon_copy_dma_ring_index(rdev);
46 break;
47 case RADEON_TEST_COPY_BLIT:
48 ring = radeon_copy_blit_ring_index(rdev);
49 break;
50 default:
51 DRM_ERROR("Unknown copy method\n");
52 return;
53 }
54
55 size = 1024 * 1024;
56
57 /* Number of tests =
58 * (Total GTT - IB pool - writeback page - ring buffers) / test size
59 */
60 n = rdev->mc.gtt_size - RADEON_IB_POOL_SIZE*64*1024;
61 for (i = 0; i < RADEON_NUM_RINGS; ++i)
62 n -= rdev->ring[i].ring_size;
63 if (rdev->wb.wb_obj)
64 n -= RADEON_GPU_PAGE_SIZE;
65 if (rdev->ih.ring_obj)
66 n -= rdev->ih.ring_size;
67 n /= size;
68
69 gtt_obj = kzalloc(n * sizeof(*gtt_obj), GFP_KERNEL);
70 if (!gtt_obj) {
71 DRM_ERROR("Failed to allocate %d pointers\n", n);
72 r = 1;
73 goto out_cleanup;
74 }
75
76 r = radeon_bo_create(rdev, size, PAGE_SIZE, true, RADEON_GEM_DOMAIN_VRAM,
77 NULL, &vram_obj);
78 if (r) {
79 DRM_ERROR("Failed to create VRAM object\n");
80 goto out_cleanup;
81 }
82 r = radeon_bo_reserve(vram_obj, false);
83 if (unlikely(r != 0))
84 goto out_cleanup;
85 r = radeon_bo_pin(vram_obj, RADEON_GEM_DOMAIN_VRAM, &vram_addr);
86 if (r) {
87 DRM_ERROR("Failed to pin VRAM object\n");
88 goto out_cleanup;
89 }
90 for (i = 0; i < n; i++) {
91 void *gtt_map, *vram_map;
92 void **gtt_start, **gtt_end;
93 void **vram_start, **vram_end;
94
95 r = radeon_bo_create(rdev, size, PAGE_SIZE, true,
96 RADEON_GEM_DOMAIN_GTT, NULL, gtt_obj + i);
97 if (r) {
98 DRM_ERROR("Failed to create GTT object %d\n", i);
99 goto out_cleanup;
100 }
101
102 r = radeon_bo_reserve(gtt_obj[i], false);
103 if (unlikely(r != 0))
104 goto out_cleanup;
105 r = radeon_bo_pin(gtt_obj[i], RADEON_GEM_DOMAIN_GTT, >t_addr);
106 if (r) {
107 DRM_ERROR("Failed to pin GTT object %d\n", i);
108 goto out_cleanup;
109 }
110
111 r = radeon_bo_kmap(gtt_obj[i], >t_map);
112 if (r) {
113 DRM_ERROR("Failed to map GTT object %d\n", i);
114 goto out_cleanup;
115 }
116
117 for (gtt_start = gtt_map, gtt_end = gtt_map + size;
118 gtt_start < gtt_end;
119 gtt_start++)
120 *gtt_start = gtt_start;
121
122 radeon_bo_kunmap(gtt_obj[i]);
123
124 if (ring == R600_RING_TYPE_DMA_INDEX)
125 r = radeon_copy_dma(rdev, gtt_addr, vram_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
126 else
127 r = radeon_copy_blit(rdev, gtt_addr, vram_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
128 if (r) {
129 DRM_ERROR("Failed GTT->VRAM copy %d\n", i);
130 goto out_cleanup;
131 }
132
133 r = radeon_fence_wait(fence, false);
134 if (r) {
135 DRM_ERROR("Failed to wait for GTT->VRAM fence %d\n", i);
136 goto out_cleanup;
137 }
138
139 radeon_fence_unref(&fence);
140
141 r = radeon_bo_kmap(vram_obj, &vram_map);
142 if (r) {
143 DRM_ERROR("Failed to map VRAM object after copy %d\n", i);
144 goto out_cleanup;
145 }
146
147 for (gtt_start = gtt_map, gtt_end = gtt_map + size,
148 vram_start = vram_map, vram_end = vram_map + size;
149 vram_start < vram_end;
150 gtt_start++, vram_start++) {
151 if (*vram_start != gtt_start) {
152 DRM_ERROR("Incorrect GTT->VRAM copy %d: Got 0x%p, "
153 "expected 0x%p (GTT/VRAM offset "
154 "0x%16llx/0x%16llx)\n",
155 i, *vram_start, gtt_start,
156 (unsigned long long)
157 (gtt_addr - rdev->mc.gtt_start +
158 (void*)gtt_start - gtt_map),
159 (unsigned long long)
160 (vram_addr - rdev->mc.vram_start +
161 (void*)gtt_start - gtt_map));
162 radeon_bo_kunmap(vram_obj);
163 goto out_cleanup;
164 }
165 *vram_start = vram_start;
166 }
167
168 radeon_bo_kunmap(vram_obj);
169
170 if (ring == R600_RING_TYPE_DMA_INDEX)
171 r = radeon_copy_dma(rdev, vram_addr, gtt_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
172 else
173 r = radeon_copy_blit(rdev, vram_addr, gtt_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
174 if (r) {
175 DRM_ERROR("Failed VRAM->GTT copy %d\n", i);
176 goto out_cleanup;
177 }
178
179 r = radeon_fence_wait(fence, false);
180 if (r) {
181 DRM_ERROR("Failed to wait for VRAM->GTT fence %d\n", i);
182 goto out_cleanup;
183 }
184
185 radeon_fence_unref(&fence);
186
187 r = radeon_bo_kmap(gtt_obj[i], >t_map);
188 if (r) {
189 DRM_ERROR("Failed to map GTT object after copy %d\n", i);
190 goto out_cleanup;
191 }
192
193 for (gtt_start = gtt_map, gtt_end = gtt_map + size,
194 vram_start = vram_map, vram_end = vram_map + size;
195 gtt_start < gtt_end;
196 gtt_start++, vram_start++) {
197 if (*gtt_start != vram_start) {
198 DRM_ERROR("Incorrect VRAM->GTT copy %d: Got 0x%p, "
199 "expected 0x%p (VRAM/GTT offset "
200 "0x%16llx/0x%16llx)\n",
201 i, *gtt_start, vram_start,
202 (unsigned long long)
203 (vram_addr - rdev->mc.vram_start +
204 (void*)vram_start - vram_map),
205 (unsigned long long)
206 (gtt_addr - rdev->mc.gtt_start +
207 (void*)vram_start - vram_map));
208 radeon_bo_kunmap(gtt_obj[i]);
209 goto out_cleanup;
210 }
211 }
212
213 radeon_bo_kunmap(gtt_obj[i]);
214
215 DRM_INFO("Tested GTT->VRAM and VRAM->GTT copy for GTT offset 0x%llx\n",
216 gtt_addr - rdev->mc.gtt_start);
217 }
218
219 out_cleanup:
220 if (vram_obj) {
221 if (radeon_bo_is_reserved(vram_obj)) {
222 radeon_bo_unpin(vram_obj);
223 radeon_bo_unreserve(vram_obj);
224 }
225 radeon_bo_unref(&vram_obj);
226 }
227 if (gtt_obj) {
228 for (i = 0; i < n; i++) {
229 if (gtt_obj[i]) {
230 if (radeon_bo_is_reserved(gtt_obj[i])) {
231 radeon_bo_unpin(gtt_obj[i]);
232 radeon_bo_unreserve(gtt_obj[i]);
233 }
234 radeon_bo_unref(>t_obj[i]);
235 }
236 }
237 kfree(gtt_obj);
238 }
239 if (fence) {
240 radeon_fence_unref(&fence);
241 }
242 if (r) {
243 printk(KERN_WARNING "Error while testing BO move.\n");
244 }
245 }
246
radeon_test_moves(struct radeon_device * rdev)247 void radeon_test_moves(struct radeon_device *rdev)
248 {
249 if (rdev->asic->copy.dma)
250 radeon_do_test_moves(rdev, RADEON_TEST_COPY_DMA);
251 if (rdev->asic->copy.blit)
252 radeon_do_test_moves(rdev, RADEON_TEST_COPY_BLIT);
253 }
254
radeon_test_create_and_emit_fence(struct radeon_device * rdev,struct radeon_ring * ring,struct radeon_fence ** fence)255 static int radeon_test_create_and_emit_fence(struct radeon_device *rdev,
256 struct radeon_ring *ring,
257 struct radeon_fence **fence)
258 {
259 int r;
260
261 if (ring->idx == R600_RING_TYPE_UVD_INDEX) {
262 r = radeon_uvd_get_create_msg(rdev, ring->idx, 1, NULL);
263 if (r) {
264 DRM_ERROR("Failed to get dummy create msg\n");
265 return r;
266 }
267
268 r = radeon_uvd_get_destroy_msg(rdev, ring->idx, 1, fence);
269 if (r) {
270 DRM_ERROR("Failed to get dummy destroy msg\n");
271 return r;
272 }
273 } else {
274 r = radeon_ring_lock(rdev, ring, 64);
275 if (r) {
276 DRM_ERROR("Failed to lock ring A %d\n", ring->idx);
277 return r;
278 }
279 radeon_fence_emit(rdev, fence, ring->idx);
280 radeon_ring_unlock_commit(rdev, ring);
281 }
282 return 0;
283 }
284
radeon_test_ring_sync(struct radeon_device * rdev,struct radeon_ring * ringA,struct radeon_ring * ringB)285 void radeon_test_ring_sync(struct radeon_device *rdev,
286 struct radeon_ring *ringA,
287 struct radeon_ring *ringB)
288 {
289 struct radeon_fence *fence1 = NULL, *fence2 = NULL;
290 struct radeon_semaphore *semaphore = NULL;
291 int r;
292
293 r = radeon_semaphore_create(rdev, &semaphore);
294 if (r) {
295 DRM_ERROR("Failed to create semaphore\n");
296 goto out_cleanup;
297 }
298
299 r = radeon_ring_lock(rdev, ringA, 64);
300 if (r) {
301 DRM_ERROR("Failed to lock ring A %d\n", ringA->idx);
302 goto out_cleanup;
303 }
304 radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore);
305 radeon_ring_unlock_commit(rdev, ringA);
306
307 r = radeon_test_create_and_emit_fence(rdev, ringA, &fence1);
308 if (r)
309 goto out_cleanup;
310
311 r = radeon_ring_lock(rdev, ringA, 64);
312 if (r) {
313 DRM_ERROR("Failed to lock ring A %d\n", ringA->idx);
314 goto out_cleanup;
315 }
316 radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore);
317 radeon_ring_unlock_commit(rdev, ringA);
318
319 r = radeon_test_create_and_emit_fence(rdev, ringA, &fence2);
320 if (r)
321 goto out_cleanup;
322
323 mdelay(1000);
324
325 if (radeon_fence_signaled(fence1)) {
326 DRM_ERROR("Fence 1 signaled without waiting for semaphore.\n");
327 goto out_cleanup;
328 }
329
330 r = radeon_ring_lock(rdev, ringB, 64);
331 if (r) {
332 DRM_ERROR("Failed to lock ring B %p\n", ringB);
333 goto out_cleanup;
334 }
335 radeon_semaphore_emit_signal(rdev, ringB->idx, semaphore);
336 radeon_ring_unlock_commit(rdev, ringB);
337
338 r = radeon_fence_wait(fence1, false);
339 if (r) {
340 DRM_ERROR("Failed to wait for sync fence 1\n");
341 goto out_cleanup;
342 }
343
344 mdelay(1000);
345
346 if (radeon_fence_signaled(fence2)) {
347 DRM_ERROR("Fence 2 signaled without waiting for semaphore.\n");
348 goto out_cleanup;
349 }
350
351 r = radeon_ring_lock(rdev, ringB, 64);
352 if (r) {
353 DRM_ERROR("Failed to lock ring B %p\n", ringB);
354 goto out_cleanup;
355 }
356 radeon_semaphore_emit_signal(rdev, ringB->idx, semaphore);
357 radeon_ring_unlock_commit(rdev, ringB);
358
359 r = radeon_fence_wait(fence2, false);
360 if (r) {
361 DRM_ERROR("Failed to wait for sync fence 1\n");
362 goto out_cleanup;
363 }
364
365 out_cleanup:
366 radeon_semaphore_free(rdev, &semaphore, NULL);
367
368 if (fence1)
369 radeon_fence_unref(&fence1);
370
371 if (fence2)
372 radeon_fence_unref(&fence2);
373
374 if (r)
375 printk(KERN_WARNING "Error while testing ring sync (%d).\n", r);
376 }
377
radeon_test_ring_sync2(struct radeon_device * rdev,struct radeon_ring * ringA,struct radeon_ring * ringB,struct radeon_ring * ringC)378 static void radeon_test_ring_sync2(struct radeon_device *rdev,
379 struct radeon_ring *ringA,
380 struct radeon_ring *ringB,
381 struct radeon_ring *ringC)
382 {
383 struct radeon_fence *fenceA = NULL, *fenceB = NULL;
384 struct radeon_semaphore *semaphore = NULL;
385 bool sigA, sigB;
386 int i, r;
387
388 r = radeon_semaphore_create(rdev, &semaphore);
389 if (r) {
390 DRM_ERROR("Failed to create semaphore\n");
391 goto out_cleanup;
392 }
393
394 r = radeon_ring_lock(rdev, ringA, 64);
395 if (r) {
396 DRM_ERROR("Failed to lock ring A %d\n", ringA->idx);
397 goto out_cleanup;
398 }
399 radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore);
400 radeon_ring_unlock_commit(rdev, ringA);
401
402 r = radeon_test_create_and_emit_fence(rdev, ringA, &fenceA);
403 if (r)
404 goto out_cleanup;
405
406 r = radeon_ring_lock(rdev, ringB, 64);
407 if (r) {
408 DRM_ERROR("Failed to lock ring B %d\n", ringB->idx);
409 goto out_cleanup;
410 }
411 radeon_semaphore_emit_wait(rdev, ringB->idx, semaphore);
412 radeon_ring_unlock_commit(rdev, ringB);
413 r = radeon_test_create_and_emit_fence(rdev, ringB, &fenceB);
414 if (r)
415 goto out_cleanup;
416
417 mdelay(1000);
418
419 if (radeon_fence_signaled(fenceA)) {
420 DRM_ERROR("Fence A signaled without waiting for semaphore.\n");
421 goto out_cleanup;
422 }
423 if (radeon_fence_signaled(fenceB)) {
424 DRM_ERROR("Fence B signaled without waiting for semaphore.\n");
425 goto out_cleanup;
426 }
427
428 r = radeon_ring_lock(rdev, ringC, 64);
429 if (r) {
430 DRM_ERROR("Failed to lock ring B %p\n", ringC);
431 goto out_cleanup;
432 }
433 radeon_semaphore_emit_signal(rdev, ringC->idx, semaphore);
434 radeon_ring_unlock_commit(rdev, ringC);
435
436 for (i = 0; i < 30; ++i) {
437 mdelay(100);
438 sigA = radeon_fence_signaled(fenceA);
439 sigB = radeon_fence_signaled(fenceB);
440 if (sigA || sigB)
441 break;
442 }
443
444 if (!sigA && !sigB) {
445 DRM_ERROR("Neither fence A nor B has been signaled\n");
446 goto out_cleanup;
447 } else if (sigA && sigB) {
448 DRM_ERROR("Both fence A and B has been signaled\n");
449 goto out_cleanup;
450 }
451
452 DRM_INFO("Fence %c was first signaled\n", sigA ? 'A' : 'B');
453
454 r = radeon_ring_lock(rdev, ringC, 64);
455 if (r) {
456 DRM_ERROR("Failed to lock ring B %p\n", ringC);
457 goto out_cleanup;
458 }
459 radeon_semaphore_emit_signal(rdev, ringC->idx, semaphore);
460 radeon_ring_unlock_commit(rdev, ringC);
461
462 mdelay(1000);
463
464 r = radeon_fence_wait(fenceA, false);
465 if (r) {
466 DRM_ERROR("Failed to wait for sync fence A\n");
467 goto out_cleanup;
468 }
469 r = radeon_fence_wait(fenceB, false);
470 if (r) {
471 DRM_ERROR("Failed to wait for sync fence B\n");
472 goto out_cleanup;
473 }
474
475 out_cleanup:
476 radeon_semaphore_free(rdev, &semaphore, NULL);
477
478 if (fenceA)
479 radeon_fence_unref(&fenceA);
480
481 if (fenceB)
482 radeon_fence_unref(&fenceB);
483
484 if (r)
485 printk(KERN_WARNING "Error while testing ring sync (%d).\n", r);
486 }
487
radeon_test_syncing(struct radeon_device * rdev)488 void radeon_test_syncing(struct radeon_device *rdev)
489 {
490 int i, j, k;
491
492 for (i = 1; i < RADEON_NUM_RINGS; ++i) {
493 struct radeon_ring *ringA = &rdev->ring[i];
494 if (!ringA->ready)
495 continue;
496
497 for (j = 0; j < i; ++j) {
498 struct radeon_ring *ringB = &rdev->ring[j];
499 if (!ringB->ready)
500 continue;
501
502 DRM_INFO("Testing syncing between rings %d and %d...\n", i, j);
503 radeon_test_ring_sync(rdev, ringA, ringB);
504
505 DRM_INFO("Testing syncing between rings %d and %d...\n", j, i);
506 radeon_test_ring_sync(rdev, ringB, ringA);
507
508 for (k = 0; k < j; ++k) {
509 struct radeon_ring *ringC = &rdev->ring[k];
510 if (!ringC->ready)
511 continue;
512
513 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", i, j, k);
514 radeon_test_ring_sync2(rdev, ringA, ringB, ringC);
515
516 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", i, k, j);
517 radeon_test_ring_sync2(rdev, ringA, ringC, ringB);
518
519 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", j, i, k);
520 radeon_test_ring_sync2(rdev, ringB, ringA, ringC);
521
522 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", j, k, i);
523 radeon_test_ring_sync2(rdev, ringB, ringC, ringA);
524
525 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", k, i, j);
526 radeon_test_ring_sync2(rdev, ringC, ringA, ringB);
527
528 DRM_INFO("Testing syncing between rings %d, %d and %d...\n", k, j, i);
529 radeon_test_ring_sync2(rdev, ringC, ringB, ringA);
530 }
531 }
532 }
533 }
534